Research On Iron Content Detection System And Blowing State Classification Of Nickel Converter Based On Machine Vision

Converter is an important pyrometallurgical equipment to realize the deep enrichment and purification of nickel sulfur,and nickel converter blowing is an important part of smelting high ice nickel in flash furnace plant of nickel smelter,which has a key influence on the yield and quality of nickel products.However,because the nickel converter blowing process is not carried out in a completely closed environment,it involves chemical reaction,heat change and transfer,fluid flow,etc.,and the material changes in a wide range,many influencing factors,the whole process has a strong dynamic change characteristics.Coupled with the vast majority of nickel smelters in China are currently operating on the nickel-sulfur blowing process only by manual experience,which not only limits the nickel converter blowing to achieve real-time detection,but also affects the judgment of the state of nickel-sulfur blowing.Therefore,the intelligence of all aspects of the nickel converter blowing process is of great practical significance to promote the high-quality development of China’s nickel smelting industry.After exploring the current development status of nickel smelting at home and abroad and the research of a nickel smelter,we deeply understand the production status and structure of nickel converter,and study the current machine vision method,image processing method,machine learning algorithm and deep learning algorithm to realize the nickel-sulfur metal-iron content detection and nickel converter blowing state identification and classification by using machine vision method.The specific work is as follows:1.In response to the problems of harsh environment in the nickel converter blowing process and the difference in the operator’s judgment of the furnace discharge time,machine vision technology is applied to develop an online nickel-sulfur metal iron content detection system.Firstly,the cross-sectional images of nickel-sulfur metal samples are collected using industrial cameras and image segmentation and preprocessing are performed;secondly,the contrast and entropy values in the image texture information are used for sample selection;finally,the image color feature information is extracted,and the linear regression-based primary model,secondary model and least squares support vector machine model are established for comparison using gray value features and actual measured iron content data,and evaluation indexes are used Evaluation was performed.The experimental results show that the least squares support vector machine model has a high detection accuracy and the detection precision meets the accuracy requirements of metal component content detection.2.In response to the traditional converter blowing status judged by manual experience,which is influenced by human subjective factors and there are many scientific problems in the judgment basis,a blowing status classification algorithm based on the channel attention mechanism is proposed,which is improved based on the Res Net18 network model,and the SENet channel attention mechanism is introduced to design the SE-Res Block module to improve the learning of image features The algorithm is based on the Res Net18 network model.Firstly,the three models of Dense Net121,Mobile Net V2 and the proposed algorithm were tested on the test set using a 10-fold cross-validation method to select appropriate hyperparameters for the models on the converter blown metal image dataset,which verified that the three models could achieve good performance under a small number of datasets;then the three models were re-trained on the whole converter blown metal image dataset using the Then the three models were retrained on the whole converter blowing metal image data set and the real prediction ability of the three models was tested on the converter blowing metal image test set.Combining the final prediction accuracy,recall rate and F1 value,it was concluded that the algorithm model performed better than Dense Net121 and Mobile Net V2,and the algorithm could classify the three blowing states of converter blowing in the early,middle and late stages with higher accuracy.